SPSSX Discussion

Tr : Overfitting in logistic regression

Classic

List

Threaded

2 messages Options

pauly vanessa-2

Tr : Overfitting in logistic regression

I've done a logistic regression with not muche events (n=3D21) for 6 variables and the reviewer ask the problem of overfitting.
Is there any method to take into account overfitting or to evaluate this in logistic regression ?
I've found that I can use cross validation techniques but I dont'know how to use it :
I do my logistic regression, I obtain my final model and then I do the cross validation?
Thanks for your help!
_______________________________________________________________________ =0A=
Envoyez avec Yahoo! Mail. Une boite mail plus intelligente http://mail.yaho=
o.fr
--0-264233468-1214562643=:30351
Content-Type: text/html; charset=utf-8
Content-Transfer-Encoding: quoted-printable

<html><head><style type=3D"text/css"></style></he=
ad><body><div style=3D"font-family:times new roman, new york, times, serif;=
font-size:12pt"><DIV>I've done a logistic regression with not muche events =
(n=3D21) for 6 variables and the reviewer ask the problem of overfitting.</=
DIV>=0A<DIV> </DIV>=0A<DIV>Is there any method to take into account ov=
erfitting or to evaluate this in logistic regression?</DIV>=0A<DIV>I've fou=
nd that I can use cross validation techniques but I dont'know how to use it=
:</DIV>=0A<DIV> </DIV>=0A<DIV>I do my logistic regression, I obtain m=
y final model and then I do the cross validation?</DIV>=0A<DIV> </DIV>=
=0A<DIV>Thanks for your help!</DIV>=0A<DIV> </DIV>=0A<DIV> </DIV>=
</div><br>=0A=0A=0A <hr size=3D"1"> =0AEnvoy=C3=A9 avec <a href=3D"htt=
p://us.rd.yahoo.com/mailuk/taglines/isp/control/*http://us.rd.yahoo.com/evt=
=3D52423/*http://fr.docs.yahoo.com/mail/overview/index.html">Yahoo! Mail</a=
>.<br>Une boite mail plus intelligente. </a></body></html>
--0-264233468-1214562643=:30351--

_____________________________________________________________________________
Envoyez avec Yahoo! Mail. Une boite mail plus intelligente http://mail.yahoo.fr

====================To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

vlad simion

Re: Tr : Overfitting in logistic regression

Hi Vanessa,

for the cross validation you compute a dummy variable that split your file
in two parts randomly and run the logistic regression on the first sample
(70% of the whole base), after that you apply the model on the other sample
(30%) and hopefully you get the same classification percentage.

compute dummy=rv.bernoulli(0.7). /* dummy will have aprox 70% values of 1 /*
exe.
logistic regression DV
/select = dummy eq 1 /* split your file in 2 samples /*
/method = enter IV1 IV2
/criteria = pin (0.05) pout (0.10) iterate (20) cut (0.5).
you might also take a look at the overdisspersion issue on Ray' site :

http://www.spsstools.net/Syntax/Distributions/FittingModelsWithOverdispersion.txt

hth,
Vlad
On Fri, Jun 27, 2008 at 1:36 PM, pauly vanessa <[hidden email]>
wrote:

>
> I've done a logistic regression with not muche events (n=3D21) for 6
> variables and the reviewer ask the problem of overfitting.
> Is there any method to take into account overfitting or to evaluate this in
> logistic regression ?
> I've found that I can use cross validation techniques but I dont'know how
> to use it :
> I do my logistic regression, I obtain my final model and then I do the
> cross validation?
> Thanks for your help!
> _______________________________________________________________________
> =0A=
> Envoyez avec Yahoo! Mail. Une boite mail plus intelligente
> http://mail.yaho=
> o.fr
> --0-264233468-1214562643=:30351
> Content-Type: text/html; charset=utf-8
> Content-Transfer-Encoding: quoted-printable
>
> <html><head><style type=3D"text/css"></style></he=
> ad><body><div style=3D"font-family:times new roman, new york, times,
> serif;=
> font-size:12pt"><DIV>I've done a logistic regression with not muche events
> =
> (n=3D21) for 6 variables and the reviewer ask the problem of
> overfitting.</=
> DIV>=0A<DIV> </DIV>=0A<DIV>Is there any method to take into account
> ov=
> erfitting or to evaluate this in logistic regression?</DIV>=0A<DIV>I've
> fou=
> nd that I can use cross validation techniques but I dont'know how to use
> it=
> :</DIV>=0A<DIV> </DIV>=0A<DIV>I do my logistic regression, I obtain m=
> y final model and then I do the cross
> validation?</DIV>=0A<DIV> </DIV>=
> =0A<DIV>Thanks for your
> help!</DIV>=0A<DIV> </DIV>=0A<DIV> </DIV>=
> </div><br>=0A=0A=0A <hr size=3D"1"> =0AEnvoy=C3=A9 avec <a
> href=3D"htt=
> p://
> us.rd.yahoo.com/mailuk/taglines/isp/control/*http://us.rd.yahoo.com/evt=
> =3D52423/*http://fr.docs.yahoo.com/mail/overview/index.html">Yahoo!
> Mail</a=
> >.<br>Une boite mail plus intelligente. </a></body></html>
> --0-264233468-1214562643=:30351--
>
>
>
>
> _____________________________________________________________________________
> Envoyez avec Yahoo! Mail. Une boite mail plus intelligente
> http://mail.yahoo.fr
>
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD